Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
Miquel dynamics for circle patterns
Sanjay Ramassamy
March 12 2018
ENS Lyon
Partly joint work with Alexey Glutsyuk (ENS Lyon & HSE)
Skoltech
Seminar of the Center for Advanced Studies
• Circle patterns form a well-studied class of objects indiscrete differential geometry (discretization of confor-mal maps).
• Many discrete integrable systems have been discoveredrecently (pentagram, dimers,...).
• Attempt to construct a discrete integrable system onsome space of circle patterns.
• First review the Goncharov-Kenyon discrete integrablesystem for dimers.
The dimer modela b
c d e
f g
Configurations :
Weights : cde bcg aef
• Probabilistic model : draw a configuration at randomwith probability proportional to its weight. Edge cor-relations ?
Dimer local moves
• Contraction of degree 2 vertices :
• Urban renewal :
1 1
a
b
c
d
1
1 1
1
∆ = ac+ bd
c/∆
d/∆ b/∆
a/∆
Goncharov-Kenyon dimer dynamics
• Start with Z2 with some edge weights. At even (resp.odd) times, perform an urban renewal on each white(resp. black) face followed by the contraction of all thedegree 2 vertices. Get Z2 with different edge weights.
Goncharov-Kenyon dimer dynamics
• Start with Z2 with some edge weights. At even (resp.odd) times, perform an urban renewal on each white(resp. black) face followed by the contraction of all thedegree 2 vertices. Get Z2 with different edge weights.
Goncharov-Kenyon dimer dynamics
• Start with Z2 with some edge weights. At even (resp.odd) times, perform an urban renewal on each white(resp. black) face followed by the contraction of all thedegree 2 vertices. Get Z2 with different edge weights.
Goncharov-Kenyon dimer dynamics
• Start with Z2 with some edge weights. At even (resp.odd) times, perform an urban renewal on each white(resp. black) face followed by the contraction of all thedegree 2 vertices. Get Z2 with different edge weights.
Goncharov-Kenyon dimer dynamics
• Start with Z2 with some edge weights. At even (resp.odd) times, perform an urban renewal on each white(resp. black) face followed by the contraction of all thedegree 2 vertices. Get Z2 with different edge weights.
m = 4
n = 2n
m
• Start with spatially biperiodic edge weights, with fun-damental domain of size m × n. The weights remainm× n biperiodic under the dynamics.
• It is seen as an m× n square grid graph on the torus.
The Kasteleyn operator K
• K is a weighted signed adjacency matrix of the graphwith edge weights.
• For planar graphs, the determinant of K gives the par-tition function (sum of the weights of all dimer config-urations). The dimer correlations are given by minorsof K−1 (Kasteleyn, Temperley, Fisher).
• For graphs on a torus, the Fourier transform of K willgive the spectral curve.
• Kasteleyn signing : assign a sign to each edge suchthat the number of minus signs around a face of degree2 mod 4 (resp. 0 mod 4) is even (resp. odd).
• We shall only consider bipartite graphs, that is graphswhere the vertices can be colored black and white witheach edge connecting a black vertex to a white vertex.
+ +
+
+ +
+−
• K : weighted signed adjacency matrix with rows (resp.columns) indexed by white (resp. black) vertices.
a bc d e
f g
+ +
+
+ +
+−
• K : weighted signed adjacency matrix with rows (resp.columns) indexed by white (resp. black) vertices.
1
2
3
1
2
3
c a 0f −d g0 b e
1 2 3
1
2
3
• Consider a bipartite graph on a torus together with twocycles generating the first homology group of the torus.
• K(z, w) : for an edge from a white vertex to a blackvertex, the corresponding coefficient is multiplied by z,z−1, w or w−1 if the edge crosses the right, left, top orbottom boundary of the fundamental domain.
1
1 2
2
a+
b+ d−
c−e+ f+
g+ h+
K(z, w) =
(b+ aw−1 h+ gz−1
f + ez −d− cw
)
Spectral data
• Spectral curve : the set of all (z, w) ∈ (C∗)2 such thatdetK(z, w) = 0.
• Divisor on the spectral curve : values of (z, w) on thespectral curve for which the first coordinate of the vec-tor vanishes.
• Associate to each (z, w) on the spectral curve a vectorin the kernel of K(z, w).
• Consider the Goncharov-Kenyon dynamics for m × nsquare-grid dimer model on the torus.
Algebro-geometric integrability
• The dynamics preserves the spectral curve, hence theintegrals of motion are the coefficients of the polynomialdetK(z, w).
• The motion of divisor corresponds to translation on theJacobian of the spectral curve.
Unrelated dimer story : Aztec diamond
Aztec diamond of size 3
Pick a dimer configurationof the Aztec diamond ofsize n uniformly at ran-dom.
picture by Cris Mooretuvalu.santafe.edu/∼moore/aztec256.gif
• Appearance of a limit shape in the limit n→∞.
color code for dimers:
• To any dimer configuration one can associate a heightfunction, whose graph is a stepped surface.
• A random dimer configuration of the Aztec diamond isa random surface.
• For n → ∞, this random surface concentrates arounda deterministic surface called the limit shape.
• The limit shape is obtained by solving a variationalproblem, minimizing a certain functional with pre-scribed boundary conditions.
• Main topic of this talk : Miquel dynamics on circlepatterns.
• Its definition by Kenyon was inspired by theGoncharov-Kenyon dimer integrable system.
Miquel’s theorem
A
A′
B′B C
DD′
C ′
Miquel’s theorem
Theorem (Miquel, 1838). In this setting,A,B,C,D concyclic ⇔ A′, B′, C ′, D′ concyclic.
A
A′
B′B C
DD′
C ′
Miquel’s theorem
A
A′
B′B C
DD′
C ′
θ1
θ2
θ3
θ4
θi : intersection anglebetween two circles
Miquel’s theorem
A
A′
B′B C
DD′
C ′
θ1
θ2
θ3
θ4
θi : intersection anglebetween two circles
Theorem (R., 2017). θ1 + θ3 = θ2 + θ4 ⇔A,B,C,D concyclic ⇔ A′, B′, C ′, D′ concyclic.
A
B
C
D
ABCD concyclic ⇔ A+ C = B + D
A
B
C
D
ABCD concyclic ⇔ A+ C = B + D
A
B
C
D
ABCD concyclic ⇔ A+ C = B + D
θ1
θ2 θ3
θ4
A = θ1 + +
C = θ3 + +
B = θ2 + +
D = θ4 + +
Square grid circle patterns
• A square grid circle pattern (SGCP) is a mapS : Z2 → R2 such that any four vertices around a faceof Z2 get mapped to four concyclic points.
S
Square grid circle patterns
• A square grid circle pattern (SGCP) is a mapS : Z2 → R2 such that any four vertices around a faceof Z2 get mapped to four concyclic points.
S
• Folds and non-convex quadrilaterals are allowed.
• Folds and non-convex quadrilaterals are allowed.
Miquel dynamics
• Checkerboard coloring of the faces of Z2 : black andwhite circles.
• Define two maps from the set of SGCPs to itself, blackmutation µB and white mutation µW .
• Black mutation µB : each vertex gets moved to theother intersection point of the two white circles it be-longs to. All the vertices move simultaneously.
• Black mutation µB : each vertex gets moved to theother intersection point of the two white circles it be-longs to. All the vertices move simultaneously.
• Black mutation µB : each vertex gets moved to theother intersection point of the two white circles it be-longs to. All the vertices move simultaneously.
• Black mutation µB : each vertex gets moved to theother intersection point of the two white circles it be-longs to. All the vertices move simultaneously.
• Black mutation µB : each vertex gets moved to theother intersection point of the two white circles it be-longs to. All the vertices move simultaneously.
• Why does µB produce an SGCP ?
• The maps µB and µW are involutions.
• Miquel dynamics : discrete-time dynamics obtained byalternating between µB and µW .
• Invented by Richard Kenyon.
Miquel’s theorem !
• The maps µB and µW are involutions.
• Miquel dynamics : discrete-time dynamics obtained byalternating between µB and µW .
• Invented by Richard Kenyon.
Biperiodic SGCPs
• An SGCP S is spatially biperiodic if there exist m,nintegers and ~u,~v ∈ R2 such that for all (x, y) ∈ R2,
S(x+m, y) = S(x, y) + ~u
S(x, y + n) = S(x, y) + ~v
m = 4
n = 2n
m
• The vector ~u (resp. ~v) is called the monodromy in thedirection (m, 0) (resp. (0, n)).
• A biperiodic SGCP is mapped by Miquel dynamics toanother biperiodic SGCP with the same periods andmonodromies.
• This reduces the problem to a finite-dimensional one.
• A biperiodic circle pattern in the plane projects downto a circle pattern on a flat torus.
[Mathematica]
Dimers vs circle patterns
• The limit shape in the dimer model is a deterministicsurface which minimizes some surface tension with pre-scribed boundary conditions (Cohn-Kenyon-Propp).
• For circle patterns, one can find the radii knowing theintersection angles by solving a variational principle.The functional minimized is similar to the one occurringfor dimers (Rivin, Bobenko-Springborn).
• Miquel dynamics mimics the Goncharov-Kenyon dimerdiscrete integrable system. Can it give us a direct con-nection between dimers and circle patterns ?
Miquel dynamics property wish list
Dimension of the space
Coordinates on that space
“Many” independent conserved quantities
Identify black and white mutation ascluster algebra mutations
Compatible Poisson bracket
Spectral curve
Miquel dynamics property wish list
Dimension of the space
Coordinates on that space
“Many” independent conserved quantities
Identify black and white mutation ascluster algebra mutations
Compatible Poisson bracket
Spectral curve
Miquel dynamics property wish list
Dimension of the space
Coordinates on that space
“Many” independent conserved quantities
Identify black and white mutation ascluster algebra mutations
Compatible Poisson bracket
Spectral curve
Miquel dynamics property wish list
Dimension of the space
Coordinates on that space
“Many” independent conserved quantities
Identify black and white mutation ascluster algebra mutations
Compatible Poisson bracket
Spectral curve
Miquel dynamics property wish list
Dimension of the space
Coordinates on that space
“Many” independent conserved quantities
Identify black and white mutation ascluster algebra mutations
Compatible Poisson bracket
Spectral curve
Miquel dynamics property wish list
Dimension of the space
Coordinates on that space
“Many” independent conserved quantities
Identify black and white mutation ascluster algebra mutations
Compatible Poisson bracket
Spectral curve
Miquel dynamics property wish list
Dimension of the space
Coordinates on that space
“Many” independent conserved quantities
Identify black and white mutation ascluster algebra mutations
Compatible Poisson bracket
Spectral curve
What space of circle patterns ?
• Fix m and n in Z+ and consider the space Mm,n ofSGCPs that have both (m, 0) and (0, n) as a period,considered up to similarity.
• SGCPs whose faces form a cell decomposition of thetorus (no folds, no non-convex quads) are an open sub-set of Mm,n.
• Bobenko-Springborn (2004) : this subspace of cell-decomposition SGCPs has dimension mn+ 1.
Coordinates forMm,n
• Four φ variables in each of the mn faces of a fundamen-tal domain.
• These variables must satisfy some relations.
Coordinates forMm,n
φN
φSφW φE
φN
φW
φS
φE
• Four φ variables in each of the mn faces of a fundamen-tal domain.
• These variables must satisfy some relations.
• Flatness at each face and vertex.
• Consistency of radii around a vertex.
φ1φ2
φ3
φ4
φ1φ2
φ3φ4 φ5
φ6
φ7
φ8
4∑i=1
φi = 2π8∑i=1
φi = 4π
sinφ12
sinφ32
sinφ52
sinφ72
sinφ22
sinφ42
sinφ62
sinφ82
= 1
m = 4
n = 2
• Flatness at each face and vertex.
• Consistency of radii around a vertex.
• Global relations across the torus, expressing theconsistency of radii and the parallelism of edges.
m = 4
n = 2
• Flatness at each face and vertex.
• Consistency of radii around a vertex.
• Global relations across the torus, expressing theconsistency of radii and the parallelism of edges.
∏sin φ
2 =∏
sin φ2
m = 4
n = 2
• Flatness at each face and vertex.
• Consistency of radii around a vertex.
• Global relations across the torus, expressing theconsistency of radii and the parallelism of edges.
∏sin φ
2 =∏
sin φ2∑
φ =∑φ
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Conjecture. One can choose mn+1 φ variables freely, theyprovide local coordinates “almost everywhere” on Mm,n.
m = 2
n = 2
imposed
deduced
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Conjecture. One can choose mn+1 φ variables freely, theyprovide local coordinates “almost everywhere” on Mm,n.
m = 2
n = 2
imposed
deduced
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Conjecture. One can choose mn+1 φ variables freely, theyprovide local coordinates “almost everywhere” on Mm,n.
m = 2
n = 2
imposed
deduced
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Conjecture. One can choose mn+1 φ variables freely, theyprovide local coordinates “almost everywhere” on Mm,n.
m = 2
n = 2
imposed
deduced
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Conjecture. One can choose mn+1 φ variables freely, theyprovide local coordinates “almost everywhere” on Mm,n.
m = 2
n = 2
imposed
deduced
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Conjecture. One can choose mn+1 φ variables freely, theyprovide local coordinates “almost everywhere” on Mm,n.
m = 2
n = 2
imposed
deduced
Theorem (R., 2017). The 4mn φ variables satisfying theabove relations coordinatize Mm,n.
We easily reconstruct the SGCP from such φ variables.
Conjecture. One can choose mn+1 φ variables freely, theyprovide local coordinates “almost everywhere” on Mm,n.
m = 2
n = 2
imposed
deduced
Local recurrence formulas
• Replace the φ variables by X = eiφ, complex numbersof modulus one.
XN
XW
XS
XE YE
YN
YW
YS
central faceis black
• How does black mutation act on the Xi’s and Yi’s ?
Y ′N = YN
1−
(1−X
−1W
)(1−Y
−1N
)(1−YW
)(1−XN
) 1−
(1−X
−1E
)(1−Y
−1N
)(1−YE
)(1−XN
) 1−
(1−XW
)(1−YN
)(1−Y
−1W
)(1−X
−1N
)1−
(1−XE
)(1−YN
)(1−Y
−1E
)(1−X
−1N
)
X ′N =
1−
(1−X
−1N
)(1−Y
−1W
)(1−Y ′N
−1)(
1−YN
)(1−XW
)(1−Y ′
W
)1−
(1−XN
)(1−YW
)(1−Y ′
N
)(1−Y
−1N
)(1−X
−1W
)(1−Y ′
W−1)
• Reminiscent of the mutation of ratios of cluster vari-ables in cluster algebras.
Conserved quantities
• The pair of monodromy vectors (~u,~v) up to similarity(two real conserved quantities).
• Signed sums of intersection angles along loops on thetorus.
• Draw dual graph to the square grid on the torus andorient the horizontal (resp. vertical) dual edges fromwhite dual vertices to black dual vertices (resp. fromblack dual vertices to white dual vertices).
m = 4
n = 2
• For any directed loop l drawn on the dual graph, defineγ(l) =
∑e∈l±θe.
• Draw dual graph to the square grid on the torus andorient the horizontal (resp. vertical) dual edges fromwhite dual vertices to black dual vertices (resp. fromblack dual vertices to white dual vertices).
m = 4
n = 2
• For any directed loop l drawn on the dual graph, defineγ(l) =
∑e∈l±θe.
• Draw dual graph to the square grid on the torus andorient the horizontal (resp. vertical) dual edges fromwhite dual vertices to black dual vertices (resp. fromblack dual vertices to white dual vertices).
m = 4
n = 2
• For any directed loop l drawn on the dual graph, defineγ(l) =
∑e∈l±θe.
• Draw dual graph to the square grid on the torus andorient the horizontal (resp. vertical) dual edges fromwhite dual vertices to black dual vertices (resp. fromblack dual vertices to white dual vertices).
m = 4
n = 2
• For any directed loop l drawn on the dual graph, defineγ(l) =
∑e∈l±θe.
• Draw dual graph to the square grid on the torus andorient the horizontal (resp. vertical) dual edges fromwhite dual vertices to black dual vertices (resp. fromblack dual vertices to white dual vertices).
m = 4
n = 2
• For any directed loop l drawn on the dual graph, defineγ(l) =
∑e∈l±θe.
minus if traversing e in the wrong way
intersection angle associated with e
• γ(l) only depends on the homology class of l on thetorus. We can upgrade γ to be a group homomorphismfrom H1(T,Z) to R/(2πZ).
Theorem (R., 2017). Black mutation and white mutationchange γ to −γ.
• Provides only two independent conserved quantities.
Isoradial patterns
• An SGCP is called isoradial if all the circles have acommon radius.
Theorem (R., 2017). Isoradial patterns are periodic pointsin Mm,n, with a common period depending on m and n.
• When (m,n) = (2, 1) or (m,n) = (4, 1), every patternis isoradial.
• The isoradial Miquel dynamics coincides with the iso-radial dimer dynamics.
Isoradial Miquel vs Goncharov-Kenyon
Isoradial SGCP withintersection anglesθe on each edge e.
Square grid withweights sin θe
2 oneach edge e.
Isoradial SGCP withintersection anglesθ′e on each edge e.
Square grid with
weights sinθ′e2 on
each edge e.
Black/whitemutation
Black/whiteG-K map
The 2× 2 case
• Construction of a 2 × 2 SGCP : pick B,D,F and Hfreely, pick E to be any point on the equilateral hyper-bola through B,D,F,H and extend it to a biperiodic
SGCP with monodromies ~u =−−→DF and ~v =
−−→BH.
A BC
D E
F
GH
I
(0, 0) (1, 0) (2, 0)
(0, 1) (1, 1) (2, 1)
(0, 2) (1, 2) (2, 2)
[Geogebra]
A BC
D E
F
GH
I
• Absolute motion : iterating Miquel dynamics, all thepoints usually drift to infinity.
A BC
D E
F
GH
I
• Relative motion : apply black or white mutation andtranslate to bring A back to its original position.
• A,C,G and I are fixed.
A BC
D E
F
GH
I
• Relative motion : apply black or white mutation andtranslate to bring A back to its original position.
• A,C,G and I are fixed.
• B,D,F and H move along arcs of circles.
Quartic curve for E
Theorem (R., 2017). E moves along the quartic curve Qdefined as the set of the points M satisfying
PM2P ′M2 − λΩM2 = k,
where λ and k are chosen such that the curve goes throughA,C,G and I.
• Given a 2× 2 SGCP, elementary construction of threepoints Ω, P, P ′ (cf Geogebra).
[Geogebra]
then
[Mathematica]
Binodal quartic curves
• As a curve in CP2, the quartic Q has two nodes, thecircular points at infinity (1 : i : 0) and (1 : −i : 0),
hence has geometric genus 1 and its normalization Q isan elliptic curve.
• Taking coordinates centered at Ω such that P is onthe horizontal axis, Q has an equation of the form
(X2 + Y 2)2 + aX2 + bY 2 + c = 0
• For any binodal quartic curve C with nodes P1 and P2,the group law on C can be defined using conics goingthrough P1, P2 and a fixed base point P0 ∈ C : theother three intersection points of the conic with C aredeclared to have zero sum.
Theorem (Glutsyuk-R., 2018). Denote by E′w (resp. E′b) therenormalized position of E after white (resp. black) mutation.
Then Miquel dynamics is translation on Q :
E′w = −E − 2A
E′b = −E − 2C
[Geogebra]
cubic then quartic
(then the end)